ChatGPT Knows Who It'll Recommend Before It Searches

suganthan.com

23 points·aworks··6 comments

The first sentence in the article immediately sounds AI written and also doesn't really make sense.

"I asked ChatGPT for the best AI note taking app. Seven words, no brand names." - The question was seven words? But later on the article says the prompt was "best AI note taking app 2026".

I don't know if there's a name for that AI writing pattern where it says something like "(Number) x, no y." but I see it all the time. It seems similar to https://tropes.fyi/tropes/negative-parallelism

Definitely. It's an inference crutch, like all the annoying uncanny features of LLM text. Models probably start doing this following reinforcement, because it's one of the nearest "paths" to staying on track with what the user wants.

"here's the honest take" (I'm switching things up so you have a less biased/overconfident answer)

"it's not x, it's y" (I don't want to get confused later about different valid interpretations of what I have so far)

and of course the original, "you're absolutely right"

Maybe read the article first?

The term I heard is "excessive contrast" or differentiation through negation. "Not X, Not Y, but Z" "No x, Just Y", "X, No Y, No W, but pure X"

Doesn't surprise me. Part of the allure of AI for the moneygrubbers is a) people trust it because ... it talks like people and b) they can inject any bias they want into it for a price.

This is like if Google didn’t show anything after sponsored results… wild, platform-undermining stuff

Reply on news.ycombinator.com